Deep learning is an important field in machine learning research. It has strong feature extraction ability and shows advanced performance in many applications including computer vision, natural language processing and speech recognition etc. Therefore, it has a wide-ranging impact on all aspects of data-driven life. However, existing studies have shown that unfairness in deep learning has increasingly damaged people’s interests in deep learning. Seeking methods that can effectively improve fairness has become one of the mainstream development directions of deep learning. This work reviews the tasks and fairness measurement methods of deep learning. In addition, we conduct experiments on typical fair deep learning datasets to implement individual fairness. The experimental results show that a balance is achieved between accuracy and fairness of classification tasks.